In brief: Jack Dorsey's Block has unveiled Dessa, an AI speech model built entirely on human conversation that can run locally on an M4 Max MacBook Pro.
Block, led by Jack Dorsey, has officially introduced an artificial intelligence speech model known as Dessa. Developed from the ground up, the system was trained exclusively on human conversation rather than traditional text datasets converted into speech.
Understanding the Dessa Speech Model
Dessa functions as a direct speech-to-speech model. Instead of translating written text into spoken audio through intermediate steps, it processes and generates spoken dialogue natively. Block released several audio samples to demonstrate the technology in action. These demonstrations feature Dessa engaging in a guessing game, debating the films of director Christopher Nolan, and having a back-and-forth dialogue with OpenAI's GPT-Live-1 voice model.
Listeners have noted that the system's conversational cadence carries the distinct conversational flavor of an awkward new friend, capturing realistic pauses, hesitations, and interpersonal dynamics found in everyday human exchanges.
Hardware Requirements and Local Execution
A notable characteristic of the newly unveiled technology is its accessibility regarding hardware. According to the announcement, Dessa is capable of running locally on a user's machine, specifically requiring a MacBook Pro equipped with an M4 Max chip or a superior setup.
Motivations Behind the Project
The development team outlines three primary drivers for building the conversational model:
- Pure technological curiosity
- Fostering better understanding between humans and artificial intelligence
- Addressing critical safety concerns
While Block highlighted safety as a core motivation for undertaking the project, the precise mechanics of how the system guarantees safety or resolves emotional safety challenges remain to be fully detailed.
Next Steps and Early Access
As the project moves forward, Block intends to distribute early access to the technology to a select, small group of specialized artificial intelligence safety researchers for further evaluation.
Conclusion
The unveiling of Dessa by Block marks a distinct entry into the conversational voice space. By utilizing a training approach built entirely on human speech and allowing local execution on high-end consumer hardware, the project opens new avenues for how voice-based artificial intelligence might operate and be evaluated in the future.
Source for this report: theverge.com.